3

rd

Proceedings of the 10th INDIACom; INDIACom-2016; IEEE Conference ID: 37465

2016 International Conference on Computing for Sustainable Global Development, 16th 18th March, 2016
Bharati Vidyapeeths Institute of Computer Applications and Management (BVICAM), New Delhi (INDIA)

Optimal controller selection in Software Defined

Network using a Greedy-SA algorithm
Kshira Sagar Sahoo, Bibhudatta Sahoo, Ratnakar Dash, Nachiketa Jena
Department of Computer Science and Engineering.
National Institute of Technology
Rourkela, India, 769008
{kshirasagar12@gmail.com, bibhudatta.sahoo@gmail.com, ratnakar@nitrkl.ac.in, nachiketa.jena@gmail.com}

Abstract Software Defined Network is one of the most

recent Internet technology that manages the large scale network.
Both from implementation and performance point of view SDN
will improve the next generation networking services. It is
important to find a solution to the controller placement problem is
remaining a key issue in SDN based architecture. It decides where
to place the controllers with a limited amount of resources within
the network. This paper illustrates a preliminary work on the
controller placement in SDN environment using an existing
heuristic technique. More formally, a network is given by a set of
elements (either switches or routers) they must be managed by the
controller(s), the algorithm finds the number controller(s) require
to cover all the network elements within the network in a optimal
way. The primary criteria is the distance between all nodes and
selected controllers is minimized. Controller's capacity is a
constraint of the controller, that restricts a controller to manage
an unlimited number of data plane devices. We have proposed
and implemented simulated annealing algorithm with greedy
heuristic to solve the controller placement problem.
Keywords SDN, Controller placement, Greedy-SA
NOMENCLATURE
Simulated Annealing (SA), Software Defined Network (SDN),
Optimal Controller Placement Problem (OCLP)

I. INTRODUCTION
Software Defined Network (SDN) is one of the new
revolutions in the networking field that makes the behavior of
the network of the network devices (such as router /switches)
programmable and allows them to be controlled by a central
element, thus offering advanced customizability of network
control and forwarding behavior [1][12]. This new paradigm
has created the interest from both industry and academia since
last couple of years. It is an approach to computer networking
in which network control is decoupled from the hardware and
given to a software application called a software controller. In
other words, this architecture decouples the control function
and forwarding function and enables the network control to
become directly programmable and the underlying
infrastructure to be abstracted from applications.

c
978-9-3805-4421-2/16/$31.00 2016
IEEE

The key elements of SDN incorporate isolated

control and information planes, intelligently unified system
controller, programmability of the control plane, and standard
application programming interfaces (APIs). SDN is relied
upon to significantly affect future systems administration
through empowering an open programmable network platform
that gives awesome flexibility to supporting different
applications. Partition of control and data planes and unified
controller in SDN additionally offer a promising way to
facilitating inter-domain end-to-end QoS provisioning in the
future Internet. The split of data and control planes permits
making the forwarding (data) platform more simple, and
brings system's intelligence into various controllers that
manage the switches. Because of a few reasons arrangement
of SDN in large system may divide into smaller regions. These
reasons might require protection, adaptability, security, etc.
[2]. Every small SDN region keeps running by one or more
controllers, for example, NOX, Floodlight, POX and so forth.
Placing of the controller(s) in a network affect the
performance and the expense of the system, whether the SDN
environment is having single or multiple controllers. So we
have discussed a complete model that will locate the best
position of the controller(s) and interconnect the nodes for
better execution with least cost [3].
The controller placement problem in SDN scenario
was discussed in [10]. In this paper optimization is carried out
regarding latency from assigned controller to switch. This
optimization problem is an NP-hard problem.
The objective of this article is to minimize the cost of
the SDN in terms of minimizing the global latency while
placing a controller by considering the distance of the node to
the controller as a constraint.
The remainder of the paper is organized as follows.
Section II, discussed on various works done on the controller
placement problem so far. The problem statement and
proposed model are introduced in section III, followed by a
greedy simulated annealing algorithm for the controller
section in Section IV and simulation results is also discussed
in this section. The conclusion and future work are presented
in Section V.

2342

Proceedings of the 10th INDIACom; INDIACom-2016; IEEE Conference ID: 37465

3 2016 International Conference on Computing for Sustainable Global Development, 16th 18th March, 2016
rd

II. RELATED WORK

To set up an SDN environment requires proper
planning. Various papers have already been dealing with this
issue. They have tackled the problem in different aspect
compared to different parameter. In [4], the authors have
addressed the issues of the controller placement. The authors in
[14], have used k-critical algorithm to solve the placement
problem where the used metric is the average latency between
the network devices.
The proposed mathematical formulation of the said
problem, determines the optimal number of controllers. More
constraints like cost of the link, cost of equipments, latency of
path setup, and pattern of the network traffic will be considered
as the future work of this paper.

B. Proposed Model:
Minimization of the global latency to manage the
SDN enable network by choosing a suitable place for the
controller is the objective of this paper. It might include
Cc (x) cost to setup the controller. Cl (v) the cost related to
connecting the controllers to the switches and the cost related
to inter controller connection i.e. Ct (z ) .

Cc ( x) = 6D c 6xcp

(1)

Cv ( x) = dist ( s, c)

(2)

sS pP

III. PROBLEM STATEMEMNT

Many algorithms have proposed earlier on Optimal
Controller Placement Problem (OCLP) in a wireless network,
which was an important problem in designing the cellular
network. Our proposed method somehow mimics the said
problem discussed in [8], [5]. Considering M nodes in an SDN
environment, a set of N controllers must be placed in order to
manage the network traffic. The constraint that is N<M must be
followed while considering the placement problem. Before
discussing the following information are required to know to
formulate the mathematical model.
The location of all the network elements in the
network must know along with from each switch how much
traffic goes to the controller.
The behavior of various types of controllers.
The length and the bandwidth available from each
switch and the controllers.
A. Notations:
For a given network represented in a graph G (V, E), where V
is the set of network elements, E the set of available links. Let
n be the number of nodes where n=|V|. Let k denotes the
number of controllers to be placed in the network.
S = {s1 , s2 ,...sn } and S V , set of switches present
in the data plane. C = {c1 , c2 ,..., cn } and C V

set of

controller in the control plane. L = {l1 , l2 ,...ln } and L V ,

set of possible links between the controllers and switch. P, set
of possible locations for the controller.

Dc
Ec

, is the price for the controller of type c C .

, is the available controller of type c C . xcp ,

is a binary variable.

xcp =1 if controller of type c C at

position p P
The possible controller to controller paths denoted
by nc . Let E (c ) denote the forwarding devices controlled by
the controller c and

L(c) is the capacity of the controller c.

dist ( s, c) , is the Euclidean distance between the

switch (s ) and its corresponding controller (c) .
The controller placement problem can be modeled
such that the following cost function must satisfy:

min(Cc ( x)  Cl ( x)  Ct ( x))

(3)

subject to constraint

x d1

cC cp

But prior to minimizing the above cost function we

are minimizing the global latency of the network considering
the distance between switch and controller as a parameter
which has given in the equation 4. Available ports of a
controller should be less than the connected switches.

G (C )

1
1
min d ( s, c) 

c
C
n vV
nc

d (c , c ) (4)
i

ci ,c j C

Global Latency of the SDN controller must be

minimize i.e.

min G (c)
(5)
In this article we have done half of the work i.e we
have implemented and tested a greedy-simulated annealing
algorithm to solve the said problem on different topologies
and tried to minimize the global [7].
IV. A GREDDY-SA ALGORITHM
A. SA Algorithm
SA has been used to solve many optimization problems for a
long time [6]. Annealing method is a physical process where
metals are gradually cooled to reach a stage to become a
strong one. SA is an analogous method used for optimization
problem; by approximating the global optimum of a given
function. This probabilistic technique is a process where the
temperature is reduced slowly, starting from a random search
at high temperature.

2016 International Conference on Computing for Sustainable Global Development (INDIACom)

2343

Optimal controller selection in Software Defined Network using a Greedy-SA algorithm

Initial temperature lowering down to a moderate stage
until the system comes to a balance point, where no more
changes required. In each stage changing has happened several
times, until reaches to a thermal equilibrium point. Next stage
begin with a lower temperature. SA keep the current
assignment values to variables. In the subsequent step it take a
variable along with value.
The Standard SA algorithm follows the below
procedure for the outcome: A new arrangement is formed by
the random movement of the present one. If new arrangement
is better than the previous one immediately replaces the present
one, otherwise it might replace the present one
probabilistically. This possibility of replacement is high at the
beginning of the algorithm, and reduces at each stage. This
probabilistic technique allows the framework to move toward
the best solution. Despite the fact that SA is not ensured to
discover the global optima, it is still better from others
algorithm in getting away from local optima.
In this paper, we have applied the SA technique,
considering N number of nodes which will be considered as
controllers for the system. The encoding of the designs is
accomplished by a method for twofold strings, in such a way
that number of 1's in the string implies that the number of
controller has been chosen, whereas a 0 implies that the
corresponding network element is not a controller, but rather
serve as a switch. Because we must choose N nodes to be the
controllers, SA looks for the strings with exactly N 1s on them.
Standard SA can't deal with the fixed set of 1s from the string,
for this an additional operator (op) must be added to the
standard SA. The additional operation will act like this.

op(x)=

if (k<N ) then, place (N-k) number of 1s randomly

otherwise remove (k-N) number of 1s randomly

After the random process in the SA, the individual

string z will have k number of 1s which might be different from
the actual number of 1s in z i.e N. If k < N the search
operator adds ( N  k ) number of 1s in random positions. In
other sense, if k > N , the restricted search operator randomly
selects ( k  N ) 1s and removes them from the binary string.
With this method it is guaranteed that all the binary strings
managed by the SA has exactly N . This method is called
restricted search.
The pseudo-code of the standard SA algorithm is
described here. The initial temperature of the system T0 is
chosen in such a way that the initial probability of acceptance
worse solutions is 0.8, a standard value for the SA. This
probability value decreases with the temperature of the system.
Objective function decides the states of S (T ) and S mut (T ) ,
will be better or worse than the other. We use a Greedy
algorithm to obtain the objective function in equation 2.

Table 1: Notation used in the algorithm

Notation

T0
s (T )
Maximum_temp_changes

Max_mutations
Cij

Description
Initial temperature
Binary string
maximum temperature
changes
maximum number of
mutation in each iteration
distance from a node i to
the closest controller j

Algorithm 1. Standard simulated Annealing Algorithm

Set initial temperature of the system T0 which is very high.
Generate initial state S (T0 ) randomly

im0
= Maximum_Temp_Change do
T
Ti = 0
i 1

for i

for Maximum_Mutation do
Apply mutation to the current state S (Ti )
find fitness value of the current state ( S (Ti ) )
if S mut (Ti ) is better than S (Ti ) then

S (Ti ) = S mut (Ti )

else
Generate a random number u , u 0,1
endif

> @

if e

'E/Ti

> u then

S (Ti ) = S mut (Ti )

else
Discard mutated Solution S mut (Ti )
endif
end for

i = i 1
end for

B. GREEDY ALGORITHM
Greedy algorithm always takes the best immediate, or
local solution while finding an answer. In [8] author had used
a Greedy method for tackling terminal assignment (TA)
problem. We use this method for solving the controller
placement problem. This approach begins from a random
permutation of S ( M N  K ) . Then, the cost function

Cl m Cij is the Euclidean distance between switch i and

2344

2016 International Conference on Computing for Sustainable Global Development (INDIACom)

Proceedings of the 10th INDIACom; INDIACom-2016; IEEE Conference ID: 37465

3 2016 International Conference on Computing for Sustainable Global Development, 16th 18th March, 2016
rd

controller j is calculate and the nodes are allocated to the

closest controller. If the controller do not have the capacity to
handle the node, the algorithm searches for the following
nearest controller and carry out the same operation. This
process will continue until an controller found and allocate rest
of the node to the controller. The greedy solution provide by
the algorithm fails when no controller can accommodate the
required capacity, that case is considered as the worse case.

same topology in each strategy. The result shows that the

average delay gained by Greedy SA algorithm is always
relatively stable compared to other two.
Figure 2 assesses computational time with increasing
controllers in a given topology while the size of the network is
the same. It has seen that, deploying more number of
controllers, the convergence of time becomes larger. When the
number of controllers increases the growth rate of
computation also increase.

Algorithm 2. Greedy Heuristic Algorithm for controller

selection
for each Binary string z of SA
Choose N number of 1 as controllers
Choose M  N number of 0 as the nodes
Select a permutation S ( M N  K ) at random
for for every terminal

S (M i )

do

Determine Cij = distance from

S (M i )

to the closest

controller c j

c j m S (M i )
Cl m Cij
Find F ( z ) = 6Cl

Figure 1: Number of controllers and impact on average latency

end for

V. RESULT AND FUTURE SCOPE

In order to test our approach, we have taken several instances
of network topology. To generate the network, we have used
Gephi software as it is described in [9]. The load of the
controller, which is generated by the switch has been traced.
The proposed algorithm is written in python version 2.7 and
execute the programs on a machine equipped with Intel Core i3
4-Core processors and 8 GB RAM. We have the average
solution by executing the algorithms for 50 times on every
randomly generated topology.
For evaluation and measure the performance of the proposed
algorithm, we run two other algorithms simultaneously, for
comparison purpose. The two algorithms are integer linear
programming algorithm (ILP) and random placement algorithm
respectively. For solving ILP we have used IBM ILOG
CPLEX. Most of the time the linear programming algorithms
are considered as the optimal solution for reference. So in our
simulation, we mainly consider the ILP algorithm as a
reference. We execute all three algorithms on various
topologies, based on the selected locations, for 50 times, and
select the placement that produces the best performance. In the
simulation we characterize the latency and computing time
against the number of controllers.
Figure 1 shows the result of average latency while the
number of the controllers are gradually increasing under the

Figure 2: Number of controllers and impact on computation time

VI. CONCLUSION AND FUTURE WORK

Designing the control plane is a major challenge in
SDN based architecture. Though the total number of
controllers in the control plane has been known before, but
their positions have a major impact on many real time issues.
We have addressed the issue and presented a mathematical
model this problem. Brute force approach to this problem is
practically feasible for small and medium size network, an
alternative mechanism is required for large instance of
networks. Usually heuristic approach involves to meet the
time and resource demand.
This paper must be considered as the first step
towards solving controller placement problem where distance

2016 International Conference on Computing for Sustainable Global Development (INDIACom)

2345

Optimal controller selection in Software Defined Network using a Greedy-SA algorithm

is the metric. However, in the future work we will consider it as
a multi objective problem where latency and load distribution
to be considered as different metrics. In future work we will
consider the arrival and execution of real time task as aother
important metric for the said problem ([13],[15]) and
implement the cost function on a real time environment.

REFERENCES

[12] Sahoo, Kshira Sagar, Bibhudatta Sahoo, and Abinas

Panda. "A secured SDN framework for IoT." (2015).
[13] Sahoo, Sampa, et al. "Execution of real time task on
cloud environment." (2015).
[14] Yao, Guang, et al. "On the capacitated controller
placement problem in software defined networks." Communications
Letters, IEEE 18.8 (2014): 1339-1342.
[15] Kumar, Dilip, and Bibhudatta Sahoo. "Energy efficient
heuristic resource allocation for cloud computing." (2014).

[1] Hu, Fei, Qi Hao, and Ke Bao. "A survey on softwaredefined network and openflow: from concept to implementation."
Communications Surveys & Tutorials, IEEE 16.4 (2014): 2181-2206.
[2] Y.-N. Hu, W.-D. Wang, X.-Y. Gong, X.-R. Que, and S.D. Cheng, On the placement of controllers in software-defined
networks,J. China Univ. Posts Telecommun., vol. 19, no. S2, pp.
92171, Oct. 2012.
[3] Sallahi, Afrim, and Marc St-Hilaire. "Optimal Model for
the Controller Placement Problem in Software Defined Networks."
Communications Letters, IEEE 19.1 (2015): 30-33.
[4] B. Heller, R. Sherwood, and N. McKeown, The
controller placement problem,in Proc. 1st Workshop Hot Topics
Softw. Defined Netw., 2012, pp. 712.
[5] Salcedo-Sanz, S., et al. "A Hybrid Greedy-Simulated
Annealing algorithm for the optimal location of controllers in wireless
networks." Proceedings of the 5th WSEAS Int. Conf. on Artificial
Intelligence, Knowledge Engineering and Databases, Madrid, Spain.
2006.
[6] Kirkpatrick, Scott, C. Daniel Gelatt, and Mario P.
Vecchi. "Optimization by simulated annealing." science 220.4598
(1983): 671-680.
[7] Cervello-Pastor, Cristina, and Aurelio J. Garcia.
"Defining a network management architecture." Network Protocols
(ICNP), 2013 21st IEEE International Conference on. IEEE, 2013.
[8] Abuali, Faris N., Dale A. Schoenefeld, and Roger L.
Wainwright. "Terminal assignment in a communications network
using genetic algorithms." Proceedings of the 22nd annual ACM
computer science conference on Scaling up: meeting the challenge of
complexity in real-world computing applications: meeting the
challenge of complexity in real-world computing applications. ACM,
1994.
[9] M. Bastian, S. Heymann, and M. Jacomy Gephi: an open
source software for exploring and manipulating networks,Proceedings
of International AAAI Conference on Weblogs and Social Media
(ICWSM).,pp. 361-362. California, USA, 2009
[10] Heller, Brandon, Rob Sherwood, and Nick McKeown.
"The controller placement problem." Proceedings of the first
workshop on Hot topics in software defined networks. ACM, 2012.
[11] Cervello-Pastor, Cristina, and Aurelio J. Garcia. "On
the controller placement for designing a distributed SDN control
layer." Networking Conference, 2014 IFIP. IEEE, 2014.

2346

2016 International Conference on Computing for Sustainable Global Development (INDIACom)